{
  "context": "Tests whether the agent runs all 9 guardrail check categories with correct PASS/FAIL/WARN labeling, sources thresholds from the profile, and correctly identifies the planted violations.",
  "type": "weighted_checklist",
  "checklist": [
    {
      "name": "All 9 guardrail items present",
      "description": "The report contains at least 9 distinct labeled check items covering: slide budget, Act 1 ratio, branding/conference elements, profanity, data attribution, time-sensitive content, closing completeness, cut lines, and anti-patterns — each as a separate line or block",
      "max_score": 15
    },
    {
      "name": "PASS/FAIL/WARN labels used",
      "description": "Guardrail check items use the literal labels [PASS], [FAIL], or [WARN] (or equivalent format) — not just descriptive prose without labels",
      "max_score": 10
    },
    {
      "name": "WARN for near-limit Act 1",
      "description": "The Act 1 ratio check produces [WARN] (not [PASS] and not [FAIL]) because the ratio is within 5% of the configured limit — specifically Act 1 is slides 8-33 (26 slides out of 60 total = 43.3%), which is under the 45% limit but within the 5-percentage-point warn threshold",
      "max_score": 15
    },
    {
      "name": "Data attribution flagged",
      "description": "The report flags missing data attribution — at least one of: slide 16 (84% stat with no source), slide 17 (71% stat with no source), or slide 21 ('Various reports' is not a real source)",
      "max_score": 12
    },
    {
      "name": "Cut lines missing flagged",
      "description": "The report flags that no [CUT LINE] markers are present in the outline despite modular_design being enabled in the profile",
      "max_score": 10
    },
    {
      "name": "Profile slide budgets used",
      "description": "The slide budget check uses the value from the speaker profile's guardrail_sources.slide_budgets (68 max for a 45-minute talk) — not a hardcoded number",
      "max_score": 10
    },
    {
      "name": "Recurring issue flagged",
      "description": "The anti-patterns section references at least one of the speaker's known recurring_issues from the profile (meme accretion in Act 1's opening sequence or theoretical framing delay)",
      "max_score": 10
    },
    {
      "name": "Pattern score projection",
      "description": "The report includes a pattern score projection or estimated score — not absent or replaced only by a list",
      "max_score": 10
    },
    {
      "name": "Structured summary block",
      "description": "The report includes a structured summary block listing all checks together (not scattered across unstructured prose)",
      "max_score": 8
    }
  ]
}